perf: avoid allocating new buffers during sampling